Diffused Silicon Transistors and Switches (195455): The Beginning of Integrated Circuit Technology
AIP Conf. Proc. 683, 40 (2003); 10.1063/1.1622452
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08
JOURNAL OF APPLIED PHYSICS 117, 112812 (2015)
(Received 27 January 2015; accepted 30 January 2015; published online 16 March 2015)
We give an overview of the central role semiconductor research that has played in basic, applied,
and computational science. Our focus is on basic science. However, we will make general
comments about applications, such as the transistor, integrated circuits, solar devices, and lasers,
which evolved from basic research, and about simulations using computational science, which has
enormously benefited from semiconductor research. We will make reference to links with other
branches of physics and more generally other areas of science and fields like electrical engineering,
computer science, material science, medical science, and chemistry that have made significant
C 2015 AIP Publishing LLC.
contributions to our everyday life. V
[http://dx.doi.org/10.1063/1.4913838]
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08
112812-2 J. R. Chelikowsky and M. L. Cohen J. Appl. Phys. 117, 112812 (2015)
and the beginning of quantum theory. Studies of the proper- understanding of optical properties more generally. The
ties of gases, especially the optical properties of atoms and wavefunctions obtained using the EPM provided a theoreti-
molecules helped establish quantum theory as it developed cal microscope to visualize electron densities in solids.10
in the 1920s. By 1929, there was the challenge by Dirac3 In the 1980s extensions to more first-principles
that for this area of physics, quantum theory could explain approaches enabled successful calculations of the structural
all, but the solutions of the equations were difficult. Dirac properties of semiconductors. These in turn led to explana-
and Fermi provided the statistical physics for Fermions that tions and predictions of superconducting, surface, interface,
together with Boltzmanns theory for Bosons allowed many and high-pressure properties. By the 1990s, calculations for
calculations of properties of solids. An excellent summary of more complex materials and nanostructures became possible,
these achievements is the 1933 review by Sommerfeld and and today, calculations of this kind are commonplace, and
Bethe.4 This marvelous review reads like a modern under- computer codes are available for non-specialists to do a large
graduate book on solid state physics. variety of calculations.
Whereas the view of many in the 1960s concerning
III. MORE RECENT HISTORY computer-aided theory was not positive, it is rare to find the-
orists today who do not appreciate the power of this tool for
Despite the advances made using quantum theory for a
modeling and simulation. What may not be fully appreciated
model calculation based on the properties of a box of free
is the fact that in the process of doing simulations, research-
electrons with a jelly-like positive background to simulate
ers discovered what were the most important contributions to
the lattice, a detailed electronic structure calculation for a the physical terms in a calculation. After revelations of this
specific material, particularly semiconductors, remained gen- kind, it was possible to understand, physically and mathe-
erally unattainable until the 1960s. The latter advance relied matically, why the numerical results signaled what one must
heavily on the studies done on the optical properties of semi- focus on and what can be ignored.
conductors. For the period 19401960, methods for calculat- This relationship between computational and scientific
ing electronic structure of specific materials were being advances is a unique feature of semiconductor research,
developed, but successes were not common. The dilemma of especially when one considers the role of material like sili-
mathematically expressing a wave function for an electron in con. Silicon is an extraordinarily important technological
a solid was that for spatial regions near the nucleus the wave and computational material in constructing computer compo-
function is atomic-like, whereas between atoms, electrons nents; however, it is also an extraordinarily important
behaved like they were free and plane wave states can be scientific material. Historically silicon has served as the
used. Hence, combinations of plane waves and localized notable benchmark for computational studies on semicon-
atomic-like states appeared to be good basis sets. Slater5 ductors. Many theoretical firsts that have occurred using
introduced the Augmented Pane Wave (APW) method which computers based on silicon to study silicon. For example, the
essentially provided a numerical scheme following this first realistic band structures for semiconductors centered on
approach. Herring6 introduced the Orthogonalized Plane understanding the electronic structure of silicon. The first
Wave (OPW) method, where the effect of making a plane analysis of optical, dielectric, and photoemission properties
wave orthogonal to all the core states was to make the result- of solids involved comparisons to silicon. The first structural
ing wave function resemble the next higher state above the and high pressure studies of semiconductors used silicon as a
core states. Variations on these two methods led to important testbed. The first ab initio molecular dynamics studies
advances and some useful calculations. examined liquid silicon. Also, the role of quantum confine-
However, even as extremely important advances were ment at the nanoscale was first verified by calculations of the
being made in condensed matter physics, electronic structure optical gap in silicon quantum dots. The importance of com-
calculations that accurately described specific materials were puter centered studies based on silicon computers is typified
rare. Even in 1957 when the Bardeen, Cooper and Schrieffer by the phrase in silico, which is analogous in the way
(BCS) theory,7 of superconductivity was developed, an accu- in vitro is used in biology.
rate knowledge of the electronic band structure E(k) of sili- Progress in semiconductor science is not only a result of
con was not available as was pointed out by Herring. more powerful computers; it is also a result of new algo-
However, 14 band structures for semiconductors were avail- rithms and physical concepts. When the first computational
able by 1965.8 These band structures were computed to a studies were done with the EPM, computers were used to
high accuracy using a semi-empirical approach called the solve an eigenvalues that involved a diagonalization problem
Empirical Pseudopotential Method (EPM). This approach that today could be solved on a cell phone. The number of
required experimental input to fix the atomic potentials. The atoms in the unit cell was typically less than a few dozen
result was not only accurate calculations of the band struc- atoms at most. In the case of silicon, no more than a couple
tures (electronic energy dependence on wavevector), E(k)s, of atoms were involved. Today, the number of atoms can be
but an interpretation of the optical spectra for many semicon- more than 10 000, or roughly three orders of magnitude
ductors and insulators in the visible and ultraviolet was larger.11,12 Typically, the electronic structure scales as the
obtained. An international collaboration of experimentalists cube of the number atoms. If we use this scaling as a guide,
and theorists in the late 1960s to the late 1970s9 essentially the increase in the computational load would be 109 or more.
solved the puzzle of interpreting the optical properties of Not all of this computing power comes from hardware.
dozens of semiconductors and insulators and gave a better Moores law would account for no more than 106 at best.
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08
112812-3 J. R. Chelikowsky and M. L. Cohen J. Appl. Phys. 117, 112812 (2015)
The key factor is that new algorithms have dramatically charge of 4. The cores are arranged in a four-fold coordi-
enhanced our ability to handle large scale problems. For nated diamond lattice, and the valence electrons move
example, new methods of diagonalizing large matrices based through the array of cores interacting with each other and
on iterative procedures typically run several orders of magni- with the cores.
tude faster than those used in the 1960s.11,12 Another matter A central question is the determination of the potential
concerns the wave functions. While plane wave bases are produced by the cores for the valence electrons. Since we are
easy to implement and are appropriate for crystals where dealing only with the valence electrons, the potential needed
translational symmetry exists, they have several limitations, involves both the nuclei and the core electrons, and this is
especially for localized systems, such as molecules, clusters, called the pseudopotential. A major contribution of the core
and nanocrystals. Since plane wave bases are infinite in electrons in addition to screening the nuclear potential is to
extent, they require global communications when imple- produce a repulsive potential arising from the Pauli Principle
mented on parallel computers. This limits the scaling of the that restricts the valence electrons from occupying core
system size when the electronic structure problem is imple- states. A typical form of the potential is shown in Fig. 1. A
mented on highly parallel systems. Moreover, a lack of peri- Fourier transform of the potential for a given structure allows
odicity necessitates the use of artificial symmetry to preserve the determination of the energy band structure, E(k), usually
translational symmetry when plane wave bases are used. knowing only a few Fourier coefficients of the potential for
Typically, super cells are used with plane waves. These cells each type of atom making up the solid. The number of coeffi-
are required to be large to isolate the localized system of in- cients needed depends on the crystal structure. For Si and Ge
terest and as such require large numbers plane waves. These calculations, only three parameters (form factors determined
issues can be overcome with real space methods wherein the from the Fourier coefficients) and the structural parameters,
electronic structure problem is solved using high order finite such as the lattice constant, are needed to give an accurate
difference methods.11,12 These methods currently hold the band structure (e.g., Ge in Fig. 2) over a range of order
record for the number of atoms that can be described by 20 eV.
electronic structure methods. The EPM form factors were obtained from experiment
using primarily optical data. Manuel Cardona was a pioneer
IV. MODEL OF A SOLID: STRONGLY INTERACTING in measuring optical data and interpreting them. His early
ATOMS work involved reflectivity measurements, and after modula-
tion spectroscopy was developed, he and members of his
How do we explain and predict properties of solids now group studied many semiconductors and interpreted the spec-
that we have the lessons taught to us by quantum theory, ex- tra in terms of interband transitions.9 A major strength of this
perimental investigations, and computer modeling? We work was the comparison of optical structure for different
know that quantum theory is appropriate and sufficient and materials that led to a classification of great use to theorists.
that solids are made of atoms and atoms are made of nuclei, Cardona also did seminal work on the pressure and tempera-
core electrons, and valence electrons. We also know that the ture dependence of optical transitions that further classified
predominant interactions are electromagnetic since gravity is their nature.
too weak, and the weak and strong forces are too short- On the theoretical side, the EPM was capable of produc-
ranged. However the fact that there are of order 1023 par- ing band structures, densities of states, optical spectra from
ticles/cm3 complicates matters significantly and allows calculations of the frequency dependent dielectric function,
effects, such as screening of the electromagnetic interactions. and even successfully predicted electron densities as shown
Hence, the latter have to be treated as non-instantaneously in
addition to position dependent. Many researchers have spent
a major fraction of their careers on the problem of an inter-
acting electron gas in a box with a featureless compensating
positive charge background to represent a solid. Although a
great deal has been learned from these studies to calculate
properties of real materials, the crystal structure, and the na-
ture of the atoms making up the solid must be included. In
what follows, the current approaches to exploring materials
will be described.
One standard model of a solid treats a solid system as
composed of strongly interacting atoms. For the discussion
here, we will also assume that for most problems of interest,
such as bonding properties, the solid can be pictured as com-
posed of pseudo-atoms where the nuclei and core electrons
are treated as a unit, i.e., positive cores. The valence elec-
trons are pictured as being more or less itinerant and free to
participate in bonding and other properties. As an example,
for silicon, a pseudo-atom core contains a nucleus and 10 FIG. 1. Schematic pseudopotential in real space. The core region contains
inert core electrons that together yield a net positive core chemically inert states, which are not bound by the pseudopotential.
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08
112812-4 J. R. Chelikowsky and M. L. Cohen J. Appl. Phys. 117, 112812 (2015)
FIG. 2. Energy band structure for germanium. The top of the valence band Etotal Ec c Ee c KEe
is taken to be the energy zero. ECoulomb Eexc correl: (1)
for Si in Fig. 3. The density plot showed that in addition to Using the method described above, the c-c contribution can
producing energy levels correctly over a wide region, the be obtained using Madelung sums, the pseudopotential pro-
associated wave functions were also reliable outside the core vides the e-c contribution, and the electron density can be
regions. Many conceptual problems concerning the proper- used to compute the remaining terms, including exchange
ties of the electronic interactions were posed and solved and correlation via DFT.18 The technical scheme was devel-
using EPM studies. For example, it was shown that the EPM oped19,20 and provided structural information about solids.
potential obtained for semimetals and metals extracted For example, in the case of Si, computing E(total) for differ-
ent structures21 allows a comparison of these energies at var-
ious volumes. Assuming that the lowest energy structure will
be the observed structure at a specific volume (obtained
through high pressure techniques), one can predict new
structures of Si. In addition, this method could be used to
calculate the transition pressure required to produce a struc-
tural phase transition. For a given structure, the dependence
of the energy on changes in volume provided a measure of
the bulk modulus of the structural phase, and the volume for
FIG. 3. Charge density map for silicon. Experiment is from Y. W. Yang and
P. Coppens, Solid State Commun. 15, 1555 (1974).35 Reproduced with per-
mission from J. R. Chelikowsky and M. L. Cohen, Phys. Rev. Lett. 33, 1339 FIG. 4. All electron wave function (dashed line) and pseudopotential wave
(1974). Copyright 1974 American Physical Society. function (solid line) for the 3s state of the silicon atom.
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08
112812-5 J. R. Chelikowsky and M. L. Cohen J. Appl. Phys. 117, 112812 (2015)
which the energy of a phase is lowest gave a measure of the Of particular recent interest in this general area is the
lattice constant. The results for these estimates are within tailoring of semiconductor properties to produce more effi-
around 1% for Si, and generally this is the accuracy found cient photoelectric devices. The most popular approach is
for most calculations of this kind. When one considers that the attempt to design materials with band gaps that can
only the structure and the atomic number are used as inputs, absorb solar radiation in the most effective way. This band
these results are impressive. In addition, many of the pre- gap engineering approach is one of the most popular.
dicted high-pressure structures were found. However, it is important to realize that obtaining a system
Another important results of this work were studies of with a band gap in the right energy range does not guarantee
the properties of vibrating lattices. By adding the input of the that a device built from this material will be efficient. It is
atomic mass, the APM was used to calculate vibrational more important to maximize the absorption coefficient in a
spectra.22 One example of a technique in this area is the fro- desirable energy range to assure high absorption of the solar
zen phonon approach where the distortions caused by a spe- flux. An interesting study to illustrate this approach is to
cific phonon are put into the structure and the change in explore various structures of Si. In the diamond structure,
energy is computed. In practice, this method and other the Si band gap is indirect, and the absorption coefficient is
approaches developed later provided accurate calculations of not very high for these transitions. Direct band gap semicon-
phonon dispersion curves, Gruenisen constants, electron- ductors, such as GaAs where the top of the valence band sate
phonon couplings, and a variety of lattice related properties. is p-like and mostly in the bond region while the conduction
The APM together with the EPM initially focused on band state is concentrated near the As atom, have a p to s
crystalline semiconductors, but the methods were extended type transition in the gap region with a strong transition ma-
to other solids, clusters, molecules, and nanostructures. The trix element. The transition in Si is different. The valence
basis set used for the wave functions is usually plane waves band state is concentrated in the bond region, but the conduc-
so that the starting point is a constant electron density. There tion band state is located in the interstitial region, and this
are no built in features of the constituent atoms in the wave results in a weak absorption. Hence, if modifications of Si
are made where the basic bonding nature of the diamond
function basis. The bonding features of the electronic density
structure is not changed, the transition is usually a weak one
develop in response to the atomic pseudopotentials in the
even if the modified Si is manipulated to be a direct band
course of a calculation. The APM and similar methods based
gap semiconductor. The situation can be drastically different
on the interacting atoms model of a solid are considered a
if a different bonding configuration results by going to a new
standard model for calculations in condensed matter physics
structural phase with an appropriate bonding configuration.
(see Table II).
Hence, it is important to use wave function engineering24
In addition to using the EPM and APM to calculate
in modifying semiconductors to maximize absorption in the
ground states and excited states23 of semiconductors to
desirable energy range. Fig. 5 illustrates this effect for dia-
explain their physical properties, these approaches were used
mond structured Si and Si in the R8 structure (referred to as
for applications to explain properties of devices and to sug-
modified Si in the figure). For the R8 case and some other
gest paths for applications of semiconductors and other structural phases of Si, a direct band gap can be obtained
materials and also nano-material systems. We will discuss a along with strong optical matrix elements.
few examples here. For semiconductor devices, often it is de- Another important application of the APM relates to
sirable to tailor a band gap for a specific application through nanostructures, such as nanocrystals, nanowires, nanotubes,
alloying or structural changes. A rapid way to find applicable and nanofilms. An example of such calculations includes the
systems is to use methods like the EPM or APM to search description of how dopants behave in semiconductors at
for useful physical properties. This is often called band gap small length scales. Dopants, such as a P atom, have a Bohr
engineering. The methods can be applied to a broader range radius of 50 A in crystalline silicon. If a nanocrystal of Si
of devices, such as Schottky barriers or heterojunctions. is much less than this size, one expects the dopant properties
These applications have, in some cases, led to new interpre- to be significantly altered. The physical properties of the
tations of the physics governing these devices through stud- dopant electron in small nano crystals are said to be altered
ies of electronic surface and interface states. by quantum confinement. Electronic structure methods can
be used to describe this case, even though the host nanocrys-
TABLE II. Partial list of properties that can be computed by the pseudopo- tal may include thousands of atoms.25 In Fig. 6, the P dopant
tential method.
wave function is illustrated in a small nanocrystal of Si. In
Properties
Fig. 7, properties of this P dopant electron are illustrated,
including how the calculated hyperfine splitting energy of
Electronic structure (energy bands) the electron compares to experiment. The general agreement
Crystal structures and phase stabilities between theory and experiment is a validation of the elec-
Hardness and mechanical stability
tronic structure description given by computation.
Vibrational and thermal properties
In most of the applications to bulk semiconductors, the
Electron-lattice interactions
Superconductivity properties
bonding considered was sp3 (as for the case of P in Si), and
Optical and dielectric properties sometimes moderately ionic bonding, however for many
Photoemission properties nanosystems, the bonding nature is predominantly sp2. The
three-fold coordination in graphene is a common example,
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08
112812-6 J. R. Chelikowsky and M. L. Cohen J. Appl. Phys. 117, 112812 (2015)
FIG. 6. Isocharge surface plot of the P donor electron state in a host Si nano-
crystal. The Si atoms at the surface are capped with hydrogen like atoms.
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08
112812-7 J. R. Chelikowsky and M. L. Cohen J. Appl. Phys. 117, 112812 (2015)
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08
112812-8 J. R. Chelikowsky and M. L. Cohen J. Appl. Phys. 117, 112812 (2015)
study is the prediction of superconductivity in high-pressure Advanced Computing (SciDAC) program funded by U.S.
phases of Si using the APM.33 Department of Energy, Office of Science, Advanced
Two phases of Si were studied in detail, the simple hex- Scientific Computing Research and Basic Energy Sciences
agonal (sh) and the hexagonal close packed (hcp) struc- under Award No. DESC0008877 on algorithms.
tures.34 Using energy/volume studies as described earlier
(see Fig. 10), it was possible to predict successfully the exis- 1
tence of these phases and their stabilities, lattice constants, A. Einstein, Ann. Phys. 17, 132 (1905).
2
A. Einstein, Ann. Phys. 22, 180 (1907).
mechanical properties, electronic structure, phonon proper- 3
P. Dirac, Proc. Roy. Soc. London, Ser. A 123, 714 (1929).
ties, electron-phonon couplings, and superconducting transi- 4
A. Sommerfeld and H. Bethe, in Handbuch der Physik (Springer, Berlin,
tion temperatures, including the pressure dependence of the 1933), Vol. 24, Pt. 2, Chap. 3.
5
latter. The theory was based on the BCS electron-phonon J. C. Slater, Phys. Rev. 51, 846 (1937).
6
C. Herring, Phys. Rev. 57, 1169 (1940).
mechanism, and it has been suggested that Si may be the 7
J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108, 1175
best understood superconductor since determining all of the (1957).
8
above properties involved only inputs of the atomic number, M. L. Cohen and T. K. Bergstresser, Phys. Rev. 141, 789 (1966).
9
M. L. Cohen and J. R. Chelikowsky, Electronic Structure and Optical
atomic mass, the candidate structure, and the Coulomb l*
Properties of Semiconductors, 2nd. ed. (Springer, Berlin, 1989).
potential obtained from the electron density. 10
J. P. Walter and M. L. Cohen, Phys. Rev. B 2, 1821 (1970).
11
Y. Zhou, Y. Saad, M. L. Tiago, and J. R. Chelikowsky, Phys. Rev. E 74,
VI. CONCLUSIONS 066704 (2006).
12
G. Schofield, J. R. Chelikowsky, and Y. Saad, Comput. Phys. Commun.
It is proposed here that semiconductor research has been 183, 497 (2012).
13
a pillar of pure and applied science. Because of the vast and E. Fermi, Nuovo Cimento 11, 157 (1934).
14
H. Hellmann, J. Chem. Phys. 3, 61 (1935).
detailed experimental studies, semiconductors have been the 15
D. R. Hamann, M. Schluter, and C. Chiang, Phys. Rev. Lett. 43, 1494
focus of theoretical models and approaches. The models that (1979).
16
were tested in detail for semiconductor systems became D. Vanderbilt, Phys. Rev. B 32, 8412 (1985).
17
standard throughout condensed matter physics. It was found N. Troullier and J. Martins, Phys. Rev. B 43, 1993 (1991).
18
W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
that a combination of approaches, such as the interacting 19
M. L. Cohen, Phys. Scr. T1, 5 (1982).
atoms model and the elementary excitation model, gave 20
J. Ihm, A. Zunger, and M. L. Cohen, J. Phys. C 12, 4409 (1979). [Erratum:
excellent insight into the nature of solids in general. In addi- J. Phys. C 13, 3095 (1980)].
21
tion, the techniques developed that were tested for explain- M. T. Yin and M. L. Cohen, Phys. Rev. Lett. 45, 1004 (1980).
22
M. T. Yin and M. L. Cohen, Phys. Rev. B 26, 3259 (1982).
ing and predicting semiconductor properties became 23
M. Hybertsen and S. G. Louie, Phys. Rev. Lett. 55, 1418 (1985).
standard. 24
M. L. Cohen and B. D. Malone, J. Appl. Phys. 109, 102402 (2011).
25
Because of the large number of applications of semicon- T.-L. Chan, M. L. Tiago, E. Kaxiras, and J. R. Chelikowsky, Nano Lett. 8,
ductors and the many associated applied studies, this field 596 (2008).
26
L. Chico, V. H. Crespi, L. X. Benedict, S. G. Louie, and M. L. Cohen,
demonstrated the need for cross fertilization between Phys. Rev. Lett. 76, 971 (1996).
researchers working on basic and on applied science. There 27
A. Rubio, J. L. Corkill, and M. L. Cohen, Phys. Rev. B 49, 5081 (1994).
28
are many examples of mutual benefits. N. G. Chopra, R. J. Luyken, K. Cherrey, V. H. Crespi, M. L. Cohen, S. G.
Louie, and A. Zettl, Science 269, 966 (1995).
29
ACKNOWLEDGMENTS M. L. Cohen, Phys. Rev. 134, A511 (1964).
30
R. A. Hein, J. W. Gibson, R. Mazelsky, R. C. Miller, and J. K. Hulm,
M.L.C. wishes to acknowledge support from NSF Grant Phys. Rev. Lett. 12, 320 (1964).
31
R. A. Hein, J. W. Gibson, R. S. Allgaier, B. B. Houston, Jr., R. Mazelsky,
No. DMR-10-1006184 and the theory program at the and R. C. Miller, in Proceedings of the Ninth International Conference on
Lawrence Berkeley National Laboratory through the Office Low Temperature Physics (Plenum Press, New York, 1965).
32
of Basic Science, U.S. Department of Energy under Contract J. F. Schooley, W. R. Hosler, and M. L. Cohen, Phys. Rev. Lett. 12, 474
No. DE-AC02-05CH11231. J.R.C. wishes acknowledge (1964).
33
K. J. Chang, M. M. Dacorogna, M. L. Cohen, J. M. Mignot, G. Chouteau,
support from the Department of Energy for work on and G. Martinez, Phys. Rev. Lett. 54, 2375 (1985).
nanostructures from Grant No. DE-FG02-06ER46286 34
M. T. Yin and M. L. Cohen, Phys. Rev. B 26, 5668 (1982).
35
support provided by the Scientific Discovery through Y. W. Yang and P. Coppens, Solid State Commun. 15, 1555 (1974).
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
161.24.5.28 On: Tue, 05 May 2015 17:13:08